ECE 435 – Network EngineeringLecture 12

Vince Weaver

http://web.eece.maine.edu/~vweaver

vincent.weaver@maine.edu

17 October 2016

http://web.eece.maine.edu/~vweaver

Announcements

• HW#5 will be posted soon

1

Midterm Review

1. OSI Layers

(a) Physical Layer – volts and bits

(b) Data Link Layer – frames

2. Physical Layer

(a) Benefit of copper over fiber

durable(?) cheaper(? depends. for local? long-

distance? for what speed?) already there(? more

details, if already there is old doesn’t help) industry

2

standard (for what?) easier to repair, bendable

(b) Benefit of fiber (typo!) over copper size, won’t corrode,

power surges attenuation (but never ”no” loss) faster.

latency (? speed of light? signals can actually travel

faster in copper than fiber)

(c) Satellite over fiber

broadcast

(d) Fiber over satellite

cheaper? Than what? for single connection, yes. For

connecting 1 million houses, maybe not.

latency more easily repaired

3

3. Ethernet

(a) vs token ring

wire breaking would take down early ethernet too

(b) collision

(c) 2.5GB existence

power-over-ethernet? Can happen with 1GB and 10GB

with injectors. Cheaper? Probably not yet. Lower

latency? Probably not. Lower power? Unclear.

4. 802.11 (typo on C/B)

(a) modify collision?

4

DCF and PCF, ack

(b) 200dBm Believable?

No, too large

5. Bluetooth

(a) Biggest problem is both use 2.4GHz, more minor is

frequency hopping interference

(b) redundant header?

simple. (power/performance?) pairing not really

relevant

6. Bridging/Switches/VLAN

5

(a) How does it handle unknown destination?

BROADCAST. Watch your wording. It broadcasts. It

may never get a response, and in that case broadcast

forever. If you say ”broadcast and wait for reply” that’s

ambiguous. It doesn’t really wait, especially not for a

”reply”. If the destination does send a frame (due to

the broadcast or otherwise) it will update the table.

(b) Reasons to partition: security, bandwidth. speed?

latency? Why might you have *higher* latency if you

partition the network?

(c) VLAN: advantage. No running to wiring closet to

6

update things, manage in software. ”easy” is relative

Expensive is also (how cheap is your intern that re-wires

things).

7. Extra Credit

(a) Bob Metcalfe

(b) Radia Perlman

7

Homework Review

1. See midterm review, many questions were reused

8

dBm Example

• Juno space probe (13 Oct 2016)8.4GHz, received −135.75dBm (2.7e-20kW) 18kb/s

9

Question from Last Time

• About order when doing Dijkstra AlgorithmI’ve updated the Lecture 11 notes

10

Tannenbaum on the Internet

• Everyone should be forced to read RFC 1958

• Top 10 principles

1. Make sure it works – do not finalize the standard until

someone has tried it first

2. Keep it simple – Occam’s razor. If feature not essential,

leave out, especially if can be achieved via other ways.

3. Make clear choices – if several ways to do something,

choose one. Having more than one way to do

11

something confuses things.

4. Exploit Modularity – protocol stacks, easy to swap out

modules independently

5. Expect heterogeneity – network should be flexible

enough to handle different hardware

6. Avoid static options and parameters – if limits are

unavoidable, best to have sender and receiver agree on

them

7. Perfect is the enemy of the good

8. Be strict when sending but tolerant when receiving –

(note, this is not always a good idea, see web browsers)

12

9. Think about scalability

10. Consider performance and cost

13

The Internet Protocol v4

• RFC791

• Network of “autonomous systems” interconnected

• Transport layer takes data and breaks into dataframesof up to 64kB. Sent through Internet (possibly broken

up) and when get to other side reconstructed by network

layer and passed up to transport layer.

• Global and unique address.Some might have multiple network ports (do they

14

have/need separate MAC addresses?)

• Need hierarchical structure to locate IP address globally

15

IPv4 Addresses

• Each IP address is 32-bits and has network address andhost ID

• Who hands these out? ICANN and various regionalauthorities Internet Corporation for Assigned Names and

Numbers

• can write many ways (all equivalent) but most commonis dotted decimal

• Classful addressing scheme

16

– Class A: 8 bit network (high bit 0). 0.0.0.0 to127.255.255.255

– Class B: 16 bit network, (high bits 10) 128.0.0.0 to191.255.255.255

– Class C: 24 bit network (high bits 110) 192.0.0.0 to223.255.255.255

– Class D: multicast (high bits 1110) 224.0.0.0 to239.255.255.255

– Class E: reserver (high bits 1111) 240.0.0.0 to255.255.255.255

• Special cases

17

– 0.0.0.0/8 reserved for local ID– 127.0.0.0/8 reserved for loopback– Internal use (192.168.x.x)– .0 represents a subnet– .1 is often (but not always) a router– it all host bits 1, broadcast for that subnet– 255.255.255.255 is broadcast for device that doesn’t

know own IP yet (DHCP)

18

Subnets

• Number of hosts available can be larger than possible

• Divide network into subnets

• All hosts on subnet have the same prefix (left bits)

• Diagram?

• Use subnet mask indicating the leftmost bits to use assubnet

19

• Can look like 255.255.255.0 meaning only bottom 8bits are for host. Alternately can write something like

192.168.8.0/24

20

CIDR address

• Running out (have run out) of network addresses

• Merge neighboring class-C together

• Scalability problem: each network takes up space inrouting table

• Solution, group neighboring class Cs together ClasslessInter-Domain Routing

• Previously, routers just shifted right for A, B, and C

21

class. Looked up A and B in table, C in hash table to

find where to send

• With CIDR bit more complex. Triplet with IP address,subnet mask, outgoing line. In theory has to scan all. If

multiple matches, one with longest mask is used. There

are algorithms to make this go faster.

• Example – from 444 in Tannenbaum

22

IP Routing

• If on same subnet, send packet directly to destination

• Otherwise, send on outgoing. See Linux route command.Often a ”default router” 0.0.0.0/0. If doesn’t match any

other, sent out over default route

• Algorithm: if to same host, skip network. If to samesubnet, deliver directly (ethernet, etc) otherwise, send

to default router

• If multiple network interfaces: If to this machine, deliver

23

it, If to directly connected subnet, directly deliver, else

deliver to next hop router

• How do we know if on network? If ((hostIP XORdestip)&subnetmask)==0

• If local, how do we map IP to MAC? We’ll see that in aminute.

• Due to CIDR, longest prefix matching. If match both a/21 and /24 then 24 is the one to send to as it’s the

longest.

24

• Data structures. Hashes? Trie?

25

IPv4 Packet Format

• Header, followed by data, multiple of 4-bytes, big-endian

• (1 nibble [4-bits]) version number: IPv4 this is 4

• (1 nibble [4-bits]) header length: variable in size. Oftenis 5 (20 bytes) the minimum, max is 15 (60 bytes)

• (1 byte) precedence/type of service/Reserved: RFC791/RFC 1349, often ignored

– Precedence (RFC 791, high bits):

26

111 (net control) 110 (internetwork control) 101

(critic/ecp) 100 (Flash override) 011 (flash) 010

(intermediate) 001 (priority) 000 (routine)

– TOS (RFC 1349):1000 minimize delay, 0100 maximize throughput, 0010:

maximize reliability, 0001 minimize cost, 0000 normal,

1111 maximize security

– R: reserved

• (2 bytes) total length (max is 64kB)

• (2 bytes) identification (ids the packet)

27

• (2 bytes) fragmentation:flags (3 bits): for fragmentation control. high bit

is always 0, next is “do not fragment” last is “more

fragments”

fragmentation offset (13-bits): all but last fragment

must be a multiple of 8-bytes as only have 13 bits to

work with)

• (1 byte) time-to-live (TTL) max routers allowed to passthough (was supposed to be time, but ended up as a

hop limit) each router decreases TTL by one, if reaches

zero discarded and ICMP error sent to source Max is

28

255. why? prevent packets from wandering lost forever

• (1 byte) upper layer protocol. RFC 1700 andwww.iana.org (ICMP=1, TCP=6, UDP=17)

• (2 bytes) header checksum, 16-bits. Sum using 16-bit1s complement, then complementing. Not as strong as

CRC-16, but faster and easier in software. Must be

recomputed each hop as TTL changes

• (4 bytes) source IP

• (4 bytes) destination IP

29

• options – not required. rare, debuggingsecurity: how secret it is (usually ignored)

strict source: gives a list of IPs of routers to traverse

loose: list of routers not to miss

record route: record ips pass on way (debugging)

timestamp(debugging)

• Data

30

IPv4 Packet Fragmentation

• Ethernet MTU 1500 but IP MTU is 64k, so must breakup larger packets

• Can be further broken up depending on MTU along way

• Final destination is responsible for reassembling

• All fragments have same sequence number. Lastfragment marked with ”more fragments” flag. Position

from fragmentation offset field

31

• Example: original, 3200 bytes of dataheader id=x, more=1, offset=0, 1480 bytes

header id=x, more=1, offset=185 1480 bytes (x8?)

header id=x, more=0, offset=370 240 bytes

• Each fragment is a valid IP packet

32

ARP – address resolution protocol

• Some way to map IP to MAC

• Could have a central server on subnet that could bequeried

• Internet uses other way. When need a mapping,broadcast to the subnet, “who has this IP”

• device reply with its IP and MAC (unicast)

• These are cached

33

• Timeout in case you reassign

• TODO: ARP Packet?

34